Data-Driven Approaches for Predicting and Forecasting Air Quality in Urban Areas DOI Creative Commons
Cosmina-Mihaela Roșca, Mădălina Cărbureanu, Adrian Stancu

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(8), С. 4390 - 4390

Опубликована: Апрель 16, 2025

Air quality (AQ) is one of the most important urban environment indicators for life. The paper proposes a software solution predicting and forecasting air index (AQI) in areas. study integrates pollutant factors (CO, NO2, SO2, PM2.5), meteorological parameters (temperature, humidity, wind speed), traffic data to determine quality. For this purpose, 19 predictive models were developed compared: 12 machine learning algorithms, 7 deep learning, 1 model based on structural component analysis. Random Forest Regression model, customized within study, achieved best results, with an R2 score 99.59%, MAE 0.22%, MAPE 0.68%, OP (Overall Precision) 95.61%. It was subsequently validated unseen recorded mean deviation 0.58%. short-term AQI (5 days), AQIF 71.62%, 0.4%, 0.9%. proposed integrated into web application IoT infrastructure real-time alert mechanisms. Future directions include expanding dataset optimizing hyperparameters increase accuracy, as well integrating PM10 O3 factors, along degree industrialization demographic level.

Язык: Английский

A Comprehensive Review of Machine Learning Models for Optimizing Wind Power Processes DOI Creative Commons
Cosmina-Mihaela Roșca, Adrian Stancu

Applied Sciences, Год журнала: 2025, Номер 15(7), С. 3758 - 3758

Опубликована: Март 29, 2025

Wind energy represents a solution for reducing environmental impact. For this reason, research studies the elements that propose optimizing wind production through intelligent solutions. Although there are address optimization of turbine performance or other indirectly related factors in production, remains topic insufficiently explored and synthesized literature. This how machine learning (ML) techniques can be applied to optimize production. aims study systematic applications ML identify analyze key stages optimized Through research, case highlighted by which methods proposed directly target issue power process turbines. From total 1049 articles obtained from Web Science database, most studied models context artificial neural networks, with 478 papers identified. Additionally, literature identifies 224 have random forest 114 incorporated gradient boosting about power. Among these, 60 specifically addressed aspect allows identification gaps The notes previous focused on forecasting, fault detection, efficiency. existing addresses indirect component performance. Thus, paper current discusses algorithms processes, future directions increasing efficiency turbines integrated predictive methods.

Язык: Английский

Процитировано

0
A Comparative Study of Azure Custom Vision Versus Google Vision API Integrated into AI Custom Models Using Object Classification for Residential Waste DOI Creative Commons
Cosmina-Mihaela Roșca, Adrian Stancu,

M. Tanase

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(7), С. 3869 - 3869

Опубликована: Апрель 1, 2025

The residential separate collection of waste is the first stage in recyclability for sustainable development. paper focuses on designing and implementing a low-cost automatic sorting bin (RBin) recycling, alleviating user’s classification burden. Next, an analysis two object identification models was conducted to sort materials into categories cardboard, glass, plastic, metal. A major challenge distinguishing between glass plastic due their similar visual characteristics. research assesses performance Azure Custom Vision Service (ACVS) model, which achieves high accuracy training data but underperforms real-time applications, with 95.13%. In contrast, second Waste Sorting Model (CWSM), demonstrates (96.25%) during proves be effective applications. CWSM uses two-tier approach, identifying descriptively using Google API (GVAS) followed by through CWSM, predicate-based custom model. employs LbfgsMaximumEntropyMulti algorithm dataset 1000 records training, divided equally across categories. This study proposes innovative evaluation metric, Weighted Classification Confidence Score (WCCS). results show that outperforms ACVS real-world testing, achieving real 99.75% after applying WCCS. explores importance customized over pre-implemented services when model characteristics not pixel-by-pixel examination.

Язык: Английский

Процитировано

0
The Impact of Cloud Versus Local Infrastructure on Automatic IoT-Driven Hydroponic Systems DOI Creative Commons
Cosmina-Mihaela Roșca, Adrian Stancu, Marian Popescu

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(7), С. 4016 - 4016

Опубликована: Апрель 5, 2025

Technological advancements in the cloud field are becoming widely used on a large scale increasing activity sectors. Agriculture is an important domain everyday life, central to human existence. This research comparatively analyzes two proposed types of infrastructures that optimize growth flow plants hydroponic system for continuous monitoring, one full-cloud and full-local. The study’s main objective determine which more suitable scenario by conducting seven tests. aims fill gap specialized literature through detailed analysis configuration, implementation methods, all implications approaches from perspective indicators. indicators response time, operational reliability, costs, configuration scalability, data accessibility, security. infrastructure uses Microsoft Azure technologies, while local variant custom-made scripts locally installed services. For both software infrastructures, hardware components identical, including M5Stack module with sensors monitoring temperature, humidity, electrical conductivity, liquid level container. test results highlight offers shorter time (200 ms compared 300 infrastructure). also showed lower costs infrastructure, making it autonomous systems. On other hand, has greater accessibility than security measures advanced. These advantages involve recurring USD 82.57/month. limitations this associated exclusion errors cybernetics simulations analysis. Another limitation concerns real short-term costs. Future will explore fluctuations long-term Additionally, studies different plant species farms be considered.

Язык: Английский

Процитировано

0
Data-Driven Approaches for Predicting and Forecasting Air Quality in Urban Areas DOI Creative Commons
Cosmina-Mihaela Roșca, Mădălina Cărbureanu, Adrian Stancu

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(8), С. 4390 - 4390

Опубликована: Апрель 16, 2025

Air quality (AQ) is one of the most important urban environment indicators for life. The paper proposes a software solution predicting and forecasting air index (AQI) in areas. study integrates pollutant factors (CO, NO2, SO2, PM2.5), meteorological parameters (temperature, humidity, wind speed), traffic data to determine quality. For this purpose, 19 predictive models were developed compared: 12 machine learning algorithms, 7 deep learning, 1 model based on structural component analysis. Random Forest Regression model, customized within study, achieved best results, with an R2 score 99.59%, MAE 0.22%, MAPE 0.68%, OP (Overall Precision) 95.61%. It was subsequently validated unseen recorded mean deviation 0.58%. short-term AQI (5 days), AQIF 71.62%, 0.4%, 0.9%. proposed integrated into web application IoT infrastructure real-time alert mechanisms. Future directions include expanding dataset optimizing hyperparameters increase accuracy, as well integrating PM10 O3 factors, along degree industrialization demographic level.

Язык: Английский

Процитировано

0